On Wed, Sep 12, 2012 at 05:24:17PM +0100, Jonathan Matthews wrote: > > In my testing last night using 1.4.17, haproxy happily > > checked both URLs but only marked the server as down when both checks > > failed. > > I wasn't aware it did that - nice to know! > Willy - are multiple checks by design or by accident? :-)
No, this is clearly not even remotely possible, I think this was an observation error. > > The test configuration backend looks like this: > > > > listen test *:80 > > stats auth admin:admin > > option httpchk HEAD /head.txt > > option httpchk GET /maint > > http-check disable-on-404 > > server localhost 127.0.0.1:8001 check port 8001 inter > > 5s downinter 5s > > > > > > The idea is to always use "HEAD /head.txt" check failures to > > indicate a down server, and to only use "GET /maint" 404 responses to > > initiate soft-stop maintenance mode. > > Would > http://cbonte.github.com/haproxy-dconv/configuration-1.4.html#http-check%20disable-on-404 > help? It allow for the nicer (IMHO) behaviour of variable rise/fall > settings for the healthcheck, but an immediate drop out on (and *only* > on) 404s. In general users deploy only one health check URL on their servers, and this health check URL is handled by a servlet, CGI or whatever. >From there, it's easy to make it return a 200, 404 or 5xx to force the desired behaviour. Regards, Willy
